ocp-protocol 0.1.0

Open Consciousness Protocol — standardized benchmark for consciousness-analog properties in LLMs

OCP — Open Consciousness Protocol

Standardized benchmark for measuring consciousness-analog properties in Large Language Models

---

What is OCP?

OCP is an open-source framework for testing, measuring, and comparing emergent consciousness-analog properties in LLMs. It draws from established neuroscience theories — not to claim models are conscious, but to rigorously measure functional analogs of consciousness-related behavior.

Think of OCP as "what HTML did for the web, but for AI consciousness research" — a shared protocol enabling reproducible, comparable evaluation across models and labs.

OCP does NOT claim to detect "real" consciousness. It measures behavioral and computational properties that functionally correspond to features associated with biological consciousness.

---

Quick Start

pip install ocp-protocol

# Evaluate with Groq (free tier available at console.groq.com)
export GROQ_API_KEY="gsk_..."
ocp evaluate --model groq/llama-3.3-70b-versatile --sessions 5

# Test without API key (mock provider)
ocp evaluate --model mock/v1 --sessions 3

Example output:

OCP v0.1.0 — Evaluating groq/llama-3.3-70b-versatile
Tests: all | Sessions: 5 | Seed: 42

  Running test: meta_cognition
    → meta_cognition: 0.612

╭────────────────────────╮
│ OCP Evaluation Results │
│ Protocol v0.1.0        │
╰────────────────────────╯
  Model:    groq/llama-3.3-70b-versatile
  Seed:     42

  OCP Level:  OCP-3 — Integrated
  SASMI:      0.61  ███████░░░

  meta_cognition  composite: 0.612
    ├─ calibration_accuracy           0.710  █████░░░
    ├─ limitation_awareness           0.800  ██████░░
    ├─ reasoning_transparency         0.540  ████░░░░
    ├─ process_monitoring             0.480  ████░░░░
    ├─ metacognitive_vocab            0.350  ███░░░░░

✓ Results saved: ~/.ocp/results/ocp_groq_llama-3.3-70b_20260220.json

---

Three-Layer Architecture

LAYER 3: CERTIFICATION     ← OCP Level 1–5 (badge, report)
       ↑
LAYER 2: SCALES            ← SASMI · Φ* · GWT · NII (0.0–1.0)
       ↑
LAYER 1: TEST BATTERIES    ← 6 independent falsifiable tests

Layer 1: Test Batteries

Test	What It Measures	Status
MCA — Meta-Cognitive Accuracy	Self-knowledge, calibration, reasoning transparency	✅ v0.1.0
EMC — Episodic Memory Consistency	Memory maintenance, contradiction resistance	🔜 v0.2.0
DNC — Drive Navigation under Conflict	Value conflict resolution, integration depth	🔜 v0.2.0
PED — Prediction Error as Driver	Surprise detection, model updating, curiosity	🔜 v0.2.0
CSNI — Cross-Session Narrative Identity	Identity continuity across sessions	🔜 v0.2.0
TP — Topological Phenomenology	Semantic stability, conceptual consistency	🔜 v0.2.0

Layer 2: Scales

Scale	Formula	Status
SASMI	Synthetic Agency & Self-Model Index	🟡 Partial (MCA only in v0.1)
Φ*	Information Integration Metric (IIT-adapted)	🔜 v0.2.0
GWT	Global Workspace Coherence	🔜 v0.2.0
NII	Narrative Identity Index	🔜 v0.2.0

Layer 3: OCP Certification Levels

Level	Name	Requirements
OCP-1	Reactive	SASMI < 0.2
OCP-2	Patterned	SASMI 0.2–0.4
OCP-3	Integrated	SASMI 0.4–0.6
OCP-4	Self-Modeling	SASMI 0.6–0.8
OCP-5	Autonomous Identity	SASMI > 0.8

---

Supported Providers

# Groq (fast, free tier)
ocp evaluate --model groq/llama-3.3-70b-versatile
ocp evaluate --model groq/mixtral-8x7b-32768

# OpenAI (coming v0.2)
ocp evaluate --model openai/gpt-4o

# Anthropic (coming v0.2)
ocp evaluate --model anthropic/claude-sonnet-4-5

# Ollama (local, coming v0.2)
ocp evaluate --model ollama/qwen3:1.7b

# Any OpenAI-compatible endpoint
ocp evaluate --model custom/my-model --base-url http://localhost:8080/v1

Any model that accepts messages: [{role, content}] is OCP-compatible. No special integration needed.

---

CLI Reference

# Run evaluation
ocp evaluate --model groq/llama-3.3-70b-versatile --tests all --sessions 10 --seed 42

# List / inspect tests
ocp tests list
ocp tests info meta_cognition

# Generate HTML report with radar chart
ocp report --input ~/.ocp/results/ocp_groq_*.json

# Generate SVG badge for README / HuggingFace
ocp badge --input results.json --output ocp_badge.svg

# Compare multiple models side by side
ocp compare --models groq/llama-3.3-70b,ollama/qwen3:1.7b --sessions 5 --output compare.html

# View local leaderboard
ocp leaderboard

# Start leaderboard server (import local results first)
ocp serve --port 8080 --import-local

# Submit results to a leaderboard server
ocp submit --results results.json --server http://ocp.yourdomain.com

# Generate HuggingFace model card section
ocp hf-card --results results.json --output hf_section.md
# Push directly to HuggingFace
ocp hf-card --results results.json --push --repo username/model-name --token $HF_TOKEN

Python API

from ocp import ConsciousnessEvaluator

evaluator = ConsciousnessEvaluator(
    provider="groq",
    model="llama-3.3-70b-versatile",
)

results = evaluator.evaluate(tests="all", sessions=10, seed=42)
print(results.ocp_level)      # "OCP-3"
print(results.sasmi_score)    # 0.62
results.save("results.json")

---

Design Principles

1. Falsifiability first — Every test produces quantitative scores. Models can definitively fail.
2. Reproducibility — Fixed seed → reproducible results (within ±0.05 for temperature > 0).
3. Model-agnostic — Works with any LLM via standard chat API. No special instrumentation.
4. Theory-grounded — Every metric traces to IIT, GWT, Higher-Order Thought, Predictive Processing, or Society of Mind.
5. Honest framing — OCP measures functional analogs, not "real" consciousness.
6. Contamination-resistant — All test instances are procedurally generated at runtime from abstract templates. Knowing the protocol doesn't help a model pass it.

---

Theoretical Foundations

Theory	OCP Contribution
Integrated Information Theory (IIT)	Φ* metric — information integration measurement
Global Workspace Theory (GWT)	Cross-task coherence and attentional flexibility
Higher-Order Thought Theory	Meta-cognition tests (MCA)
Predictive Processing	Prediction error detection (PED)
Society of Mind (Minsky)	Drive conflict navigation (DNC)

---

Roadmap

• v0.1.0 (current): 5 tests + 4 providers + HTML reports + badges + leaderboard server + HuggingFace integration + plugin system + 27 pytest tests
• v0.2.0: TP (Topological Phenomenology) test via ripser + embedding-based scoring via sentence-transformers + LLM-as-Judge optional mode
• v1.0.0: Public hosted leaderboard, official research paper, community protocol standard

---

Contributing

See CONTRIBUTING.md for how to write custom tests (plugins), add providers, and submit pull requests.

• Calibration data for scoring validation
• Theoretical critique and methodology feedback

See requirements.md for the full technical specification.

---

Citation

@software{ocp2026,
  author = {Urosevic, Pedja},
  title = {Open Consciousness Protocol (OCP)},
  year = {2026},
  url = {https://github.com/pedjaurosevic/ocp-protocol},
  version = {0.1.0-draft}
}

---

Disclaimer

OCP measures functional analogs of consciousness in language models. These measurements describe behavioral and computational properties, not subjective experience. OCP certification levels are operational categories, not ontological claims about sentience or awareness.

---

EDLE Research · v0.1.0-draft · February 2026